Quasi-greedy bases for sequences with gaps (Q2019394)

Greedy algorithms are currently in much use in various fields, for example approximations by kernel methods and methods of deep learning and neural networks. They and in particular the threshold greedy algorithms use expansions of elements of Banach spaces and approximate them by choosing large (a largest) coefficient(s) times the associated basis functions as an approximant. The methods then work recursively, but it is not yet possible to take care of expansion series whose indices have (large) gaps. This is the goal of this particular article. Examples are provided by trigonometric series -- so-called lacunary trigonometric series [\textit{A. Zygmund}, Trigonometric series. Vols. 1, 2. 2nd ed. Cambridge: At the University Press (1959; Zbl 0085.05601)]. The authors prove and generalise a number of results that were posed in the recent literature. Here is an example: it was not yet known whether a sequence \(n_1(x) < n_2(x) < n_3(x) < \cdots\) exists for an \(x\) in a Banach space with Markushevich basis such that the greedy sums \(G_{n_k}\) converge to \(x\). In this paper it is shown that it does not. Another negative answer for Schauder bases is that there are no \(\mathbf n\)-quasi greedy bases that are at the same time not quasi-greedy such that \(\mathbf n\) does not have arbitrarily large gaps. (Here \(\mathbf n\) is a sequence of \(n_i\), \(i\) a positive integer, with arbitrarily large gaps.)